export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/utils:$PATH:/sbin:/usr/sbin
ONLY=${ONLY:-"$*"}
-# bug number for skipped test:
+# bug number for skipped test: 3815 3939
+ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36 40"
+# bug number for skipped test:4178 4176
+ALWAYS_EXCEPT="$ALWAYS_EXCEPT 200 221 223b 31a"
# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
-# skip test cases failed before landing - Jinshan
-
-ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 31a 34 35 36"
-ALWAYS_EXCEPT="$ALWAYS_EXCEPT 200 201 221 223a 223b 225"
LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
copytool_cleanup() {
trap - EXIT
local agents=${1:-$(facet_active_host $SINGLEAGT)}
+ local mdtno
+ local idx
+ local oldstate
+ local mdt_hsmctrl
do_nodesv $agents "pkill -INT -x $HSMTOOL_BASE" || return 0
sleep 1
echo "Copytool is stopped on $agents"
+
+ # clean all CDTs orphans requests from previous tests
+ # that would otherwise need to timeout to clear.
+ for mdtno in $(seq 1 $MDSCOUNT); do
+ idx=$(($mdtno - 1))
+ mdt_hsmctrl="mdt.$FSNAME-MDT000${idx}.hsm_control"
+ oldstate=$(do_facet mds${mdtno} "$LCTL get_param -n " \
+ "$MDT_HSMCTRL")
+ # skip already stop[ed,ing] CDTs
+ echo $oldstate | grep stop || continue
+
+ do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=shutdown"
+ wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
+ "stopped" 20 ||
+ error "mds${mdtno} cdt state is not stopped"
+ do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=$oldstate"
+ wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
+ "$oldstate" 20 ||
+ error "mds${mdtno} cdt state is not $oldstate"
+ done
}
copytool_suspend() {
copytool_remove_backend() {
local fid=$1
- local be=$(find $HSM_ARCHIVE -name $fid)
+ local be=$(do_facet $SINGLEAGT find $HSM_ARCHIVE -name $fid)
echo "Remove from backend: $fid = $be"
do_facet $SINGLEAGT rm -f $be
}
}
cleanup_large_files() {
- local ratio=$(df $MOUNT |awk '{print $5}' |sed 's/%//g' |grep -v Use)
+ local ratio=$(df -P $MOUNT | tail -1 | awk '{print $5}' |
+ sed 's/%//g')
[ $ratio -gt 50 ] && find $MOUNT -size +10M -exec rm -f {} \;
}
local request=$2
local state=$3
- local cmd="$LCTL get_param -n $HSM_PARAM.agent_actions"
+ local cmd="$LCTL get_param -n $HSM_PARAM.actions"
cmd+=" | awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
wait_result $SINGLEMDS "$cmd" $state 100 ||
local fid=$1
local request=$2
- do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.agent_actions |"\
+ do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
"awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
}
local fid=$1
local request=$2
- do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.agent_actions |"\
+ do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
"awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
}
wait_all_done() {
local timeout=$1
- local cmd="$LCTL get_param -n $HSM_PARAM.agent_actions"
+ local cmd="$LCTL get_param -n $HSM_PARAM.actions"
cmd+=" | egrep 'WAITING|STARTED'"
wait_result $SINGLEMDS "$cmd" "" $timeout ||
check_hsm_flags $f "0x0000000d"
local LSZ=$(stat -c "%s" $f)
- local ASZ=$(stat -c "%s" $HSM_ARCHIVE/$tdir/$tfile)
+ local ASZ=$(do_facet $SINGLEAGT stat -c "%s" $HSM_ARCHIVE/$tdir/$tfile)
echo "Verifying imported size $LSZ=$ASZ"
[[ $LSZ -eq $ASZ ]] || error "Incorrect size $LSZ != $ASZ"
CURR_FILE="$CURR_DIR/$tfile.$f"
# write file-specific data
do_facet $SINGLEAGT \
- echo "d=$d, f=$f, dir=$CURR_DIR, "\
- "file=$CURR_FILE" > $CURR_FILE
+ "echo d=$d, f=$f, dir=$CURR_DIR, "\
+ "file=$CURR_FILE > $CURR_FILE"
done
done
# import to Lustre
$LFS hsm_archive $file
wait_request_state $fid ARCHIVE SUCCEED
- $LFS hsm_release $file ||
- check_hsm_flags $file "0x0000000d"
+ $LFS hsm_release $file
+ check_hsm_flags $file "0x0000000d"
$LFS hsm_restore $file
wait_request_state $fid RESTORE SUCCEED
$RUNAS $LFS hsm_state $file ||
error "user '$RUNAS_ID' cannot get HSM state of '$file'"
- $LFS hsm_release $file ||
- check_hsm_flags $file "0x0000000d"
+ $LFS hsm_release $file
+ check_hsm_flags $file "0x0000000d"
# Check that ordinary user can accessed released file.
sum1=$($RUNAS md5sum $file) ||
}
run_test 24b "root can archive, release, and restore user files"
+cleanup_test_24c() {
+ trap 0
+ set_hsm_param user_request_mask RESTORE
+ set_hsm_param group_request_mask RESTORE
+ set_hsm_param other_request_mask RESTORE
+}
+
+test_24c() {
+ local file=$DIR/$tdir/$tfile
+ local action=archive
+ local user_save
+ local group_save
+ local other_save
+
+ # test needs a running copytool
+ copytool_setup
+
+ mkdir -p $DIR/$tdir
+
+ # Save the default masks and check that cleanup_24c will
+ # restore the request masks correctly.
+ user_save=$(get_hsm_param user_request_mask)
+ group_save=$(get_hsm_param group_request_mask)
+ other_save=$(get_hsm_param other_request_mask)
+
+ [ "$user_save" == RESTORE ] ||
+ error "user_request_mask is '$user_save' expected 'RESTORE'"
+ [ "$group_save" == RESTORE ] ||
+ error "group_request_mask is '$group_save' expected 'RESTORE'"
+ [ "$other_save" == RESTORE ] ||
+ error "other_request_mask is '$other_save' expected 'RESTORE'"
+
+ trap cleanup_test_24c EXIT
+
+ # User.
+ rm -f $file
+ make_small $file
+ chown $RUNAS_ID:nobody $file ||
+ error "cannot chown '$file' to '$RUNAS_ID:nobody'"
+
+ set_hsm_param user_request_mask ""
+ $RUNAS $LFS hsm_$action $file &&
+ error "$action by user should fail"
+
+ set_hsm_param user_request_mask $action
+ $RUNAS $LFS hsm_$action $file ||
+ error "$action by user should succeed"
+
+ # Group.
+ rm -f $file
+ make_small $file
+ chown nobody:$RUNAS_GID $file ||
+ error "cannot chown '$file' to 'nobody:$RUNAS_GID'"
+
+ set_hsm_param group_request_mask ""
+ $RUNAS $LFS hsm_$action $file &&
+ error "$action by group should fail"
+
+ set_hsm_param group_request_mask $action
+ $RUNAS $LFS hsm_$action $file ||
+ error "$action by group should succeed"
+
+ # Other.
+ rm -f $file
+ make_small $file
+ chown nobody:nobody $file ||
+ error "cannot chown '$file' to 'nobody:nobody'"
+
+ set_hsm_param other_request_mask ""
+ $RUNAS $LFS hsm_$action $file &&
+ error "$action by other should fail"
+
+ set_hsm_param other_request_mask $action
+ $RUNAS $LFS hsm_$action $file ||
+ error "$action by other should succeed"
+
+ copytool_cleanup
+ cleanup_test_24c
+}
+run_test 24c "check that user,group,other request masks work"
+
+cleanup_test_24d() {
+ trap 0
+ mount -o remount,rw $MOUNT2
+}
+
+test_24d() {
+ local file1=$DIR/$tdir/$tfile
+ local file2=$DIR2/$tdir/$tfile
+ local fid1
+ local fid2
+
+ copytool_setup
+
+ mkdir -p $DIR/$tdir
+ rm -f $file1
+ fid1=$(make_small $file1)
+
+ trap cleanup_test_24d EXIT
+
+ mount -o remount,ro $MOUNT2
+
+ fid2=$(path2fid $file2)
+ [ "$fid1" == "$fid2" ] ||
+ error "FID mismatch '$fid1' != '$fid2'"
+
+ $LFS hsm_archive $file2 &&
+ error "archive should fail on read-only mount"
+ check_hsm_flags $file1 "0x00000000"
+
+ $LFS hsm_archive $file1
+ wait_request_state $fid1 ARCHIVE SUCCEED
+
+ $LFS hsm_release $file1
+ $LFS hsm_restore $file2
+ wait_request_state $fid1 RESTORE SUCCEED
+
+ $LFS hsm_release $file1 || error "cannot release '$file1'"
+ dd if=$file2 of=/dev/null bs=1M || "cannot read '$file2'"
+
+ $LFS hsm_release $file2 &&
+ error "release should fail on read-only mount"
+
+ copytool_cleanup
+ cleanup_test_24d
+}
+run_test 24d "check that read-only mounts are respected"
+
test_25a() {
# test needs a running copytool
copytool_setup
}
run_test 30b "Restore at exec (release case)"
+test_30c() {
+ needclients 2 || return 0
+
+ # test needs a running copytool
+ copytool_setup
+
+ mkdir -p $DIR/$tdir
+ local f=$DIR/$tdir/SLEEP
+ local fid=$(copy_file /bin/sleep $f)
+ chmod 755 $f
+ $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
+ wait_request_state $fid ARCHIVE SUCCEED
+ $LFS hsm_release $f
+ check_hsm_flags $f "0x0000000d"
+ # set no retry action mode
+ cdt_set_no_retry
+ do_node $CLIENT2 "$f 10" &
+ local pid=$!
+ sleep 3
+ echo 'Hi!' > $f
+ [[ $? == 0 ]] && error "Update during exec of released file must fail"
+ wait $pid
+ [[ $? == 0 ]] || error "Execution failed during run"
+ cmp /bin/sleep $f
+ [[ $? == 0 ]] || error "Binary overwritten during exec"
+
+ # cleanup
+ # remove no try action mode
+ cdt_clear_no_retry
+ check_hsm_flags $f "0x00000009"
+
+ copytool_cleanup
+}
+run_test 30c "Update during exec of released file must fail"
+
restore_and_check_size() {
local f=$1
local fid=$2
test_100() {
double_verify_reset_hsm_param loop_period
double_verify_reset_hsm_param grace_delay
- double_verify_reset_hsm_param request_timeout
+ double_verify_reset_hsm_param active_request_timeout
double_verify_reset_hsm_param max_requests
double_verify_reset_hsm_param default_archive_id
}
echo "Current requests"
local res=$(do_facet $SINGLEMDS "$LCTL get_param -n\
- $HSM_PARAM.agent_actions |\
+ $HSM_PARAM.actions |\
grep -v CANCELED | grep -v SUCCEED | grep -v FAILED")
[[ -z "$res" ]] || error "Some request have not been canceled"
cdt_disable
$LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f
local data1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
- $HSM_PARAM.agent_actions |\
+ $HSM_PARAM.actions |\
grep $fid | cut -f16 -d=")
cdt_enable
$LFS hsm_archive $DIR/$tdir/$i
done
local reqcnt1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
- $HSM_PARAM.agent_actions |\
+ $HSM_PARAM.actions |\
grep WAITING | wc -l")
cdt_restart
cdt_disable
local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
- $HSM_PARAM.agent_actions |\
+ $HSM_PARAM.actions |\
grep WAITING | wc -l")
cdt_enable
cdt_purge
}
run_test 227 "changelog when explicit setting of HSM flags"
+test_228() {
+ # test needs a running copytool
+ copytool_setup
+
+ dd if=/dev/urandom of=$DIR/$tfile bs=1M count=1 conv=sync ||
+ error "creating $DIR/$tfile"
+ $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tfile
+ wait_request_state $(path2fid $DIR/$tfile) ARCHIVE SUCCEED
+
+ $LFS hsm_release $DIR/$tfile
+ check_hsm_flags $DIR/$tfile "0x0000000d"
+
+ filefrag $DIR/$tfile | grep " 1 extent found" ||
+ error "filefrag on released file must return only one extent"
+
+ # only newer versions of cp detect sparse files by stat/FIEMAP
+ # (LU-2580)
+ cp --sparse=auto $DIR/$tfile $DIR/$tfile.2 ||
+ error "copying $DIR/$tfile"
+ cmp $DIR/$tfile $DIR/$tfile.2 || error "comparing copied $DIR/$tfile"
+
+ $LFS hsm_release $DIR/$tfile
+ check_hsm_flags $DIR/$tfile "0x0000000d"
+
+ mkdir $DIR/$tdir
+
+ tar cf - --sparse $DIR/$tfile | tar xvf - -C $DIR/$tdir ||
+ error "tar failed"
+ cmp $DIR/$tfile $DIR/$tdir/$DIR/$tfile ||
+ error "comparing untarred $DIR/$tfile"
+
+ copytool_cleanup
+}
+run_test 228 "On released file, return extend to FIEMAP. For [cp,tar] --sparse"
+
test_250() {
# test needs a running copytool
copytool_setup
while [[ $cnt != 0 || $wt != 0 ]]; do
sleep 1
cnt=$(do_facet $SINGLEMDS "$LCTL get_param -n\
- $HSM_PARAM.agent_actions |\
+ $HSM_PARAM.actions |\
grep STARTED | grep -v CANCEL | wc -l")
[[ $cnt -le $maxrequest ]] ||
error "$cnt > $maxrequest too many started requests"
wt=$(do_facet $SINGLEMDS "$LCTL get_param\
- $HSM_PARAM.agent_actions |\
+ $HSM_PARAM.actions |\
grep WAITING | wc -l")
echo "max=$maxrequest started=$cnt waiting=$wt"
done
cdt_disable
# to have a short test
- local old_to=$(get_hsm_param request_timeout)
- set_hsm_param request_timeout 4
+ local old_to=$(get_hsm_param active_request_timeout)
+ set_hsm_param active_request_timeout 4
# to be sure the cdt will wake up frequently so
# it will be able to cancel the "old" request
local old_loop=$(get_hsm_param loop_period)
sleep 5
wait_request_state $fid ARCHIVE CANCELED
- set_hsm_param request_timeout $old_to
+ set_hsm_param active_request_timeout $old_to
set_hsm_param loop_period $old_loop
copytool_cleanup
git://git.whamcloud.com / fs / lustre-release.git / blobdiff